A Rolling RFM-based Prediction Model of Online Repurchase Behavior: A Case of Women's Apparel at Yahoo! Taiwan Auction Website

• Main Authors: Chih-han Yu, 余芷函
• Other Authors: Chin-Yuan Ho
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/5ea8e8

碩士 === 國立中央大學 === 資訊管理學系 === 102 === Online shopping has grown rapidly so that B2C e-commerce gets more attention by both practitioners and researchers. While the seller has more opportunities to reach more online consumers, the online shopper has more choices as well. By focusing on returning customers, online sellers can increase revenues in a more cost-effective way. To realize the potential profits, online sellers need an efficient and effective prediction tool to capture their customers’ purchase behavior. Targeting on the woman apparel at Yahoo! Taiwan auction website, this study uses the real transaction data to develop a rolling prediction model of the online repurchase behavior, which exhibits both stability and prediction accuracy. The dataset collected from Yahoo! Taiwan auction website includes all transaction data dated before September 30, 2013 and the total number of transaction records is over 5.58 million. Based on this rich dataset, we applied a comprehensive description statistics to observe characteristics of repeat customers. We also propose a rolling repurchase behavior prediction model with up to six independent variables, including RFM (recency, frequency, total/average monetary), the last rating and the number of repurchased sellers. Classification rates of different time points and time intervals used in prediction were examined to validate the model. Through tests of goodness of model fit and logistic regression analysis, we found that the recency and the average monetary are negatively related to the probability of repurchase, whereas the higher the frequency, the total monetary, the last rating, and the number of repurchased sellers, the repurchase is more likely to occur. Only the result of the number of repurchased sellers is contradictory to our hypothesis. The contribution of this study has three: (1) practically help online sellers with target marketing to retain old customers; (2) augment the RFM model with the last rating and the number of repurchased sellers can enhance prediction accuracy effectively; (3) the description statistics based on all real transactions can be a reference for online shoppers’ behavior research.